Commit Graph

1023 Commits

Author SHA1 Message Date
Andrew Trick 10ffc2b6c2 Various bits of framework needed for precise machine-level selection
DAG scheduling during isel. Most new functionality is currently
guarded by -enable-sched-cycles and -enable-sched-hazard.

Added InstrItineraryData::IssueWidth field, currently derived from
ARM itineraries, but could be initialized differently on other targets.

Added ScheduleHazardRecognizer::MaxLookAhead to indicate whether it is
active, and if so how many cycles of state it holds.

Added SchedulingPriorityQueue::HasReadyFilter to allowing gating entry
into the scheduler's available queue.

ScoreboardHazardRecognizer now accesses the ScheduleDAG in order to
get information about it's SUnits, provides RecedeCycle for bottom-up
scheduling, correctly computes scoreboard depth, tracks IssueCount, and
considers potential stall cycles when checking for hazards.

ScheduleDAGRRList now models machine cycles and hazards (under
flags). It tracks MinAvailableCycle, drives the hazard recognizer and
priority queue's ready filter, manages a new PendingQueue, properly
accounts for stall cycles, etc.

llvm-svn: 122541
2010-12-24 05:03:26 +00:00
Andrew Trick c416ba612b whitespace
llvm-svn: 122539
2010-12-24 04:28:06 +00:00
Chris Lattner 11a33811b6 flags -> glue for selectiondag
llvm-svn: 122509
2010-12-23 17:24:32 +00:00
Chris Lattner f647e95b9a sdisel flag -> glue.
llvm-svn: 122507
2010-12-23 17:13:18 +00:00
Chris Lattner 3e5fbd74ed rename MVT::Flag to MVT::Glue. "Flag" is a terrible name for
something that just glues two nodes together, even if it is
sometimes used for flags.

llvm-svn: 122310
2010-12-21 02:38:05 +00:00
Chris Lattner 17f906be96 improve "cannot yet select" errors a trivial amount: now
they are just as useless, but at least a bit more gramatical

llvm-svn: 122305
2010-12-21 02:07:03 +00:00
Chris Lattner 77a8a71414 fix PR8642: if a critical edge has a PHI value that can trap,
isel is *required* to split the edge.  PHI values get evaluated
on the edge, not in their predecessor block.

llvm-svn: 122170
2010-12-19 04:58:57 +00:00
Bill Wendling 54df187f25 Check for _setjmp too, because it's also used.
llvm-svn: 119875
2010-11-20 00:03:09 +00:00
Duncan Sands 1462777017 Simplify uses of MVT and EVT. An MVT can be compared directly
with a SimpleValueType, while an EVT supports equality and
inequality comparisons with SimpleValueType.

llvm-svn: 118169
2010-11-03 12:17:33 +00:00
Nick Lewycky 90b2ac2696 For statistics that are only used in functions declared in !NDEBUG, wrap the
declarations in !NDEBUG to avoid -Wunused-variable warnings. Patch by
Matt Beaumont-Gay!

llvm-svn: 117345
2010-10-26 00:51:57 +00:00
Devang Patel 43c3f4b63c Simplify.
Do not count use of sdisel for single call instruction.

llvm-svn: 117316
2010-10-25 21:31:46 +00:00
Devang Patel 3bc6d198fb Add counters to count basic blocks and machine basic blocks with out of order line number info.
Add counters to count how many basic blocks are entirely selected by fastisel.

llvm-svn: 117310
2010-10-25 20:55:43 +00:00
Chandler Carruth 82058c05f8 Move the remaining attribute macros to systematic names based on the attribute
name and prefixed with 'LLVM_'.

llvm-svn: 117203
2010-10-23 08:40:19 +00:00
Owen Anderson 6c18d1aac0 Get rid of static constructors for pass registration. Instead, every pass exposes an initializeMyPassFunction(), which
must be called in the pass's constructor.  This function uses static dependency declarations to recursively initialize
the pass's dependencies.

Clients that only create passes through the createFooPass() APIs will require no changes.  Clients that want to use the
CommandLine options for passes will need to manually call the appropriate initialization functions in PassInitialization.h
before parsing commandline arguments.

I have tested this with all standard configurations of clang and llvm-gcc on Darwin.  It is possible that there are problems
with the static dependencies that will only be visible with non-standard options.  If you encounter any crash in pass
registration/creation, please send the testcase to me directly.

llvm-svn: 116820
2010-10-19 17:21:58 +00:00
Jakob Stoklund Olesen 665aa6efcc When isel is emitting instructions for an x86 target without CMOV, the CFG is
edited during emission.

If the basic block ends in a switch that gets lowered to a jump table, any
phis at the default edge were getting updated wrong. The jump table data
structure keeps a pointer to the header blocks that wasn't getting updated
after the MBB is split.

This bug was exposed on 32-bit Linux when disabling critical edge splitting in
codegen prepare.

The fix is to uipdate stale MBB pointers whenever a block is split during
emission.

llvm-svn: 115191
2010-09-30 19:44:31 +00:00
Oscar Fuentes 57214f533a Fix VS 2010 build.
Patch by Nathan Jeffords!

llvm-svn: 114661
2010-09-23 16:59:36 +00:00
Chris Lattner a9e57e0eff Rework passing parent pointers into complexpatterns, I forgot
that complex patterns are matched after the entire pattern has
a structural match, therefore the NodeStack isn't in a useful
state when the actual call to the matcher happens.

llvm-svn: 114489
2010-09-21 22:00:25 +00:00
Devang Patel 99ff76212a If only user of a vreg is an copy instruction to export copy of vreg out of current basic block then insert DBG_VALUE so that debug value of the variable is also transfered to new vreg.
Testcase is in r114476.
This fixes radar 8412415.

llvm-svn: 114478
2010-09-21 20:56:33 +00:00
Chris Lattner 0bb8b19865 correct this logic.
llvm-svn: 114474
2010-09-21 20:46:40 +00:00
Chris Lattner dd83548fea just like they can opt into getting the root of the pattern being
matched, allow ComplexPatterns to opt into getting the parent node
of the operand being matched.

llvm-svn: 114472
2010-09-21 20:37:12 +00:00
Chris Lattner eeba0c73e5 implement rdar://6653118 - fastisel should fold loads where possible.
Since mem2reg isn't run at -O0, we get a ton of reloads from the stack,
for example, before, this code:

int foo(int x, int y, int z) {
  return x+y+z;
}

used to compile into:

_foo:                                   ## @foo
	subq	$12, %rsp
	movl	%edi, 8(%rsp)
	movl	%esi, 4(%rsp)
	movl	%edx, (%rsp)
	movl	8(%rsp), %edx
	movl	4(%rsp), %esi
	addl	%edx, %esi
	movl	(%rsp), %edx
	addl	%esi, %edx
	movl	%edx, %eax
	addq	$12, %rsp
	ret

Now we produce:

_foo:                                   ## @foo
	subq	$12, %rsp
	movl	%edi, 8(%rsp)
	movl	%esi, 4(%rsp)
	movl	%edx, (%rsp)
	movl	8(%rsp), %edx
	addl	4(%rsp), %edx    ## Folded load
	addl	(%rsp), %edx     ## Folded load
	movl	%edx, %eax
	addq	$12, %rsp
	ret

Fewer instructions and less register use = faster compiles.

llvm-svn: 113102
2010-09-05 02:18:34 +00:00
Chris Lattner f6418b804e zap dead code.
llvm-svn: 112155
2010-08-26 02:57:35 +00:00
Owen Anderson a7aed18624 Reapply r110396, with fixes to appease the Linux buildbot gods.
llvm-svn: 110460
2010-08-06 18:33:48 +00:00
Owen Anderson bda59bd247 Revert r110396 to fix buildbots.
llvm-svn: 110410
2010-08-06 00:23:35 +00:00
Owen Anderson 755aceb5d0 Don't use PassInfo* as a type identifier for passes. Instead, use the address of the static
ID member as the sole unique type identifier.  Clean up APIs related to this change.

llvm-svn: 110396
2010-08-05 23:42:04 +00:00
Evan Cheng 37b740c4bf Add an ILP scheduler. This is a register pressure aware scheduler that's
appropriate for targets without detailed instruction iterineries.
The scheduler schedules for increased instruction level parallelism in
low register pressure situation; it schedules to reduce register pressure
when the register pressure becomes high.

On x86_64, this is a win for all tests in CFP2000. It also sped up 256.bzip2
by 16%.

llvm-svn: 109300
2010-07-24 00:39:05 +00:00
Gabor Greif dde79d8f1a mass elimination of reliance on automatic iterator dereferencing
llvm-svn: 109103
2010-07-22 13:36:47 +00:00
Dan Gohman 103c4ebea5 Use the source-order scheduler instead of the "fast" scheduler at -O0,
because it's more likely to keep debug line information in its original
order.

llvm-svn: 108496
2010-07-16 02:01:19 +00:00
Dan Gohman d7b5ce3312 Reapply bottom-up fast-isel, with several fixes for x86-32:
- Check getBytesToPopOnReturn().
 - Eschew ST0 and ST1 for return values.
 - Fix the PIC base register initialization so that it doesn't ever
   fail to end up the top of the entry block.

llvm-svn: 108039
2010-07-10 09:00:22 +00:00
Bill Wendling f831d86311 Clarify what mysterious check means.
llvm-svn: 108005
2010-07-09 19:44:12 +00:00
Bob Wilson 6586e9b203 --- Reverse-merging r107947 into '.':
U    utils/TableGen/FastISelEmitter.cpp
--- Reverse-merging r107943 into '.':
U    test/CodeGen/X86/fast-isel.ll
U    test/CodeGen/X86/fast-isel-loads.ll
U    include/llvm/Target/TargetLowering.h
U    include/llvm/Support/PassNameParser.h
U    include/llvm/CodeGen/FunctionLoweringInfo.h
U    include/llvm/CodeGen/CallingConvLower.h
U    include/llvm/CodeGen/FastISel.h
U    include/llvm/CodeGen/SelectionDAGISel.h
U    lib/CodeGen/LLVMTargetMachine.cpp
U    lib/CodeGen/CallingConvLower.cpp
U    lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
U    lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
U    lib/CodeGen/SelectionDAG/FastISel.cpp
U    lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
U    lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
U    lib/CodeGen/SelectionDAG/InstrEmitter.cpp
U    lib/CodeGen/SelectionDAG/TargetLowering.cpp
U    lib/Target/XCore/XCoreISelLowering.cpp
U    lib/Target/XCore/XCoreISelLowering.h
U    lib/Target/X86/X86ISelLowering.cpp
U    lib/Target/X86/X86FastISel.cpp
U    lib/Target/X86/X86ISelLowering.h

llvm-svn: 107987
2010-07-09 16:37:18 +00:00
Dan Gohman 0b5aa1cdd3 Re-apply bottom-up fast-isel, with fixes. Be very careful to avoid emitting
a DBG_VALUE after a terminator, or emitting any instructions before an EH_LABEL.

llvm-svn: 107943
2010-07-09 00:39:23 +00:00
Bill Wendling a992445ff2 Extension of r107506. Make sure that we don't mark a function as having a call
if the inline ASM doesn't need a stack frame.

llvm-svn: 107922
2010-07-08 22:38:02 +00:00
Dan Gohman e75704369d Revert 107840 107839 107813 107804 107800 107797 107791.
Debug info intrinsics win for now.

llvm-svn: 107850
2010-07-08 01:00:56 +00:00
Dan Gohman 1adc499dda Fix -fast-isel-abort to check the right instruction.
llvm-svn: 107839
2010-07-07 23:47:25 +00:00
Dan Gohman e7ccc51cc1 Implement bottom-up fast-isel. This has the advantage of not requiring
a separate DCE pass over MachineInstrs.

llvm-svn: 107804
2010-07-07 19:20:32 +00:00
Dan Gohman b792f844ad Update the insert position after scheduling, which may change the
position when emitting multiple blocks when executing a custom
inserter.

llvm-svn: 107797
2010-07-07 18:22:13 +00:00
Dan Gohman ffe64b1ee5 Give FunctionLoweringInfo an MBB member, avoiding the need to pass it
around everywhere, and also give it an InsertPt member, to enable isel
to operate at an arbitrary position within a block, rather than just
appending to a block.

llvm-svn: 107791
2010-07-07 16:47:08 +00:00
Dan Gohman 87fb4e8fcd Simplify FastISel's constructor by giving it a FunctionLoweringInfo
instance, rather than pointers to all of FunctionLoweringInfo's
members.

This eliminates an NDEBUG ABI sensitivity.

llvm-svn: 107789
2010-07-07 16:29:44 +00:00
Dan Gohman e784616fbb Move FunctionLoweringInfo.h out into include/llvm/CodeGen. This will
allow target-specific fast-isel code to make use of it directly.

llvm-svn: 107787
2010-07-07 16:01:37 +00:00
Dan Gohman ee0cb70381 CanLowerReturn doesn't need a SelectionDAG; it just needs an LLVMContext.
SelectBasicBlock doesn't needs its BasicBlock argument.

llvm-svn: 107712
2010-07-06 22:19:37 +00:00
Bill Wendling f844642350 Proper indentation.
llvm-svn: 107581
2010-07-04 08:58:43 +00:00
Dale Johannesen 4d887f7ca7 Propagate the AlignStack bit in InlineAsm's to the
PrologEpilog code, and use it to determine whether
the asm forces stack alignment or not.  gcc consistently
does not do this for GCC-style asms; Apple gcc inconsistently
sometimes does it for asm blocks.  There is no
convenient place to put a bit in either the SDNode or
the MachineInstr form, so I've added an extra operand
to each; unlovely, but it does allow for expansion for
more bits, should we need it.  PR 5125.  Some
existing testcases are affected.
The operand lists of the SDNode and MachineInstr forms
are indexed with awesome mnemonics, like "2"; I may
fix this someday, but not now.  I'm not making it any
worse.  If anyone is inspired I think you can find all
the right places from this patch.

llvm-svn: 107506
2010-07-02 20:16:09 +00:00
Dan Gohman 93f5920914 Rename CreateReg to CreateRegs, and MakeReg to CreateReg.
llvm-svn: 107451
2010-07-02 00:10:16 +00:00
Dan Gohman d2965c10a1 Temporarily disable on-demand fast-isel.
llvm-svn: 107393
2010-07-01 12:15:30 +00:00
Dan Gohman 42b7ee15f5 Use FuncInfo's isExportedInst accessor method instead of
doing the work manually.

llvm-svn: 107384
2010-07-01 03:57:05 +00:00
Dan Gohman 85e02e9340 Rename CreateRegForValue to CreateReg, and change its argument
from a Value to a Type, because it doesn't actually care about
the Value.

llvm-svn: 107383
2010-07-01 03:55:39 +00:00
Dan Gohman 722f5fc567 Enable on-demand fast-isel.
llvm-svn: 107377
2010-07-01 02:58:57 +00:00
Dan Gohman 600f62b3ba Reapply r106634, now that the bug it exposed is fixed.
llvm-svn: 106746
2010-06-24 14:30:44 +00:00
Daniel Dunbar 4df321b7ad Revert r106263, "Fold the ShrinkDemandedOps pass into the regular DAGCombiner pass,"... it was causing both 'file' (with clang) and 176.gcc (with llvm-gcc) to be miscompiled.
llvm-svn: 106634
2010-06-23 17:09:26 +00:00